Neurobiology of Disease

Reactive Astrogliosis Causes the Development ofSpontaneous Seizures

Stefanie Robel,1 X Susan C. Buckingham,1 X Jessica L. Boni,1 Susan L. Campbell,1 Niels C. Danbolt,2

Therese Riedemann,3 Bernd Sutor,3 and Harald Sontheimer1

1Department of Neurobiology, Center for Glial Biology in Medicine, University of Alabama at Birmingham, Birmingham, Alabama 35209, 2Department ofAnatomy, Institute of Basic Medical Sciences, University of Oslo, 0317 Oslo, Norway, and 3Institute of Physiology, Department of Physiological Genomics,Ludwig-Maximilians-University of Munich, 80336 Munich, Germany

Epilepsy is one of the most common chronic neurologic diseases, yet approximately one-third of affected patients do not respond toanticonvulsive drugs that target neurons or neuronal circuits. Reactive astrocytes are commonly found in putative epileptic foci and havebeen hypothesized to be disease contributors because they lose essential homeostatic capabilities. However, since brain pathologyinduces astrocytes to become reactive, it is difficult to distinguish whether astrogliosis is a cause or a consequence of epileptogenesis. Wenow present a mouse model of genetically induced, widespread chronic astrogliosis after conditional deletion of �1-integrin (Itg�1). Inthese mice, astrogliosis occurs in the absence of other pathologies and without BBB breach or significant inflammation. Electroenceph-alography with simultaneous video recording revealed that these mice develop spontaneous seizures during the first six postnatal weeksof life and brain slices show neuronal hyperexcitability. This was not observed in mice with neuronal-targeted �1-integrin deletion,supporting the hypothesis that astrogliosis is sufficient to induce epileptic seizures. Whole-cell patch-clamp recordings from astrocytesfurther suggest that the heightened excitability was associated with impaired astrocytic glutamate uptake. Moreover, the relative expres-sion of the cation-chloride cotransporters (CCC) NKCC1 (Slc12a2) and KCC2 (Slc12a5), which are responsible for establishing theneuronal Cl � gradient that governs GABAergic inhibition were altered and the NKCC1 inhibitor bumetanide eliminated seizures in asubgroup of mice. These data suggest that a shift in the relative expression of neuronal NKCC1 and KCC2, similar to that observed inimmature neurons during development, may contribute to astrogliosis-associated seizures.

Key words: astrocytes; astrogliosis; bumetanide; EEG; epilepsy; seizures

IntroductionEpilepsy is the most prevalent chronic brain disorder (Thurmanet al., 2011) characterized by repetitive, spontaneous seizures.Seizures are clinical manifestations of hypersynchronous neuro-nal activity originating at restricted sites of the brain. From theseepileptogenic foci, the activity can spread to larger portions of thecerebral cortex resulting in neuronal network hyperexcitabilitycaused by an imbalance between synaptic excitation and synapticinhibition (Jefferys, 1990; Pinto et al., 2005; Trevelyan et al., 2006;Carmignoto and Haydon, 2012). Almost all available therapeutic

anticonvulsants either act on synaptic mechanisms to restore thebalance between excitation and inhibition or on voltage-dependention channels to reduce neuronal excitability. However, approxi-mately one-third of the patients affected suffer from therapy-resistant epilepsies (Wahab et al., 2010). It seems to be a legitimateassumption that disturbances of synaptic mechanisms and/orchanges in neuronal excitability are epiphenomena of other possiblymore severe dysfunctions. Therefore, a better understanding of theunderlying mechanisms leading to epileptic seizures is essential.

It is now widely acknowledged that most neuropathologiesinvolve both neuronal and glial cell dysfunction. Astrocytes havecritical roles in the healthy brain, including the maintenance ofion and neurotransmitter homeostasis (Volterra and Meldolesi,2005). Changes in astrocyte morphology, biochemistry, and physi-ology in response to CNS insults are collectively termed “reactiveastrogliosis” (Sofroniew and Vinters, 2010).

Astrogliosis is also thought to contribute to the epilepsy dis-ease phenomenon (Olsen and Sontheimer, 2008; Wetheringtonet al., 2008; Seifert et al., 2010). In animal models, reactive astro-cytes undergo extensive physiological changes that alter ion andneurotransmitter homeostasis, and intracellular and extracellu-lar water content, which can cause neuronal hyperexcitability(Bordey and Sontheimer, 1998; Bordey et al., 2001; Seifert andSteinhauser, 2011).

Received April 17, 2014; revised Dec. 16, 2014; accepted Jan. 12, 2015.Author contributions: S.R., S.C.B., S.L.C., and H.S. designed research; S.R., S.C.B., J.L.B., S.L.C., T.R., and B.S.

performed research; N.C.D. contributed unpublished reagents/analytic tools; S.R., S.C.B., J.L.B., S.L.C., T.R., and B.S.analyzed data; S.R. and B.S. wrote the paper.

This work was supported by National Institutes of Health (NIH) grants 2RO1NS036692, 5RO1NS031234,5R01NS052634, 1F31NS074597, and 1R01NS082851. S.R. received funding from the German Research Foundation,the Epilepsy Foundation, and the American Brain Tumor Association. S.C.B. directs the EEG Core funded by NationalInstitutes of Health Eunice Kennedy Shriver Intellectual and Developmental Disabilities #5P30HD038985. We thankBriana D. Miller and Miki Jinno for assistance with the mouse colony and for technical support. We are grateful toLeda Dimou for supplying �1-integrin, Nex::Cre mice.

The authors declare no competing financial interests.Correspondence should be addressed to Stefanie Robel, 1719 6th Avenue South, CIRC 252C, Birmingham, AL

35294. E-mail: srobel@uab.edu.DOI:10.1523/JNEUROSCI.1574-14.2015

Copyright © 2015 the authors 0270-6474/15/353330-16$15.00/0

3330 • The Journal of Neuroscience, February 25, 2015 • 35(8):3330 –3345

Does astrogliosis precede neuronal network hyperexcitability,or does astrogliosis result from seizures and then exacerbate hy-perexcitability? Ortinski et al. (2010) investigated the role of as-trogliosis in the development of hyperexcitability after adenoviralinfection of hippocampal astrocytes. This led to downregulationof the astrocytic enzyme glutamine synthetase, which transformsglutamate into glutamine. Loss of this enzyme impaired the exci-tation/inhibition balance of network activity within the hip-pocampus, leading to neuronal hyperexcitability in acute slices.

In our study, we investigated whether reactive astrogliosiscauses epileptic seizures in vivo. For this purpose, we condition-ally deleted �1-integrin (Itg�1) in radial glia, which results inchronic reactive astrogliosis in almost every region of the brainwithout gross brain abnormalities or pronounced inflammation(Robel et al., 2009). We show that these animals develop neuronalhyperexcitability and spontaneous recurrent seizures. This mightbe attributable in part to reduced expression of the chloridetransporter KCC2, which counteracts neuronal chloride importof NKCC1. Balanced expression of NKCC1 and KCC2 are crucialfor the proper function of the inhibitory neurotransmitterGABA. Inhibition of NKCC1 with bumetanide rescued the sei-zure phenotype in a subset of mice. Patch-clamp recordings inacute brain slices show reduced astrocytic Kir currents and im-paired glutamate homeostasis, which together may provide anenhanced excitatory drive. Importantly, neuronal hyperexcitabil-ity was not observed in mice with neuron-specific �1-integrindeletion. Thus, genetically induced astrogliosis is sufficient toinvoke neuronal changes that produce hyperexcitability and ep-ileptic seizures, albeit the mechanism whereby astrogliosis causesthese neuronal changes is not known.

Materials and MethodsAnimals. FVB-N//129/Sv-�1-integrin mice carrying alleles for Itg�1flanked by loxP sites �1-integrin fl/fl (Potocnik et al., 2000) were mated tomice expressing a Cre-recombinase under the hGFAP promoter(hGFAP-Cre �) (FVB-Tg (GFAP-cre) 25Mes/J; The Jackson Laboratory,strain #004600) to recombine in astrocytes and neurons (Zhuo et al.,2001), or in neurons only when expressed under the Nex promoter(Nex::Cre �/�; Goebbels et al., 2006). To genetically label astrocytes,FVB-N//Swiss Webster-Aldh1l1-eGFP mice (Cahoy et al., 2008) weremated to the �1-integrin//hGFAP-Cre mouse line. Offspring was usedafter three generations of backcrossing �1-integrin//hGFAP-Cre mice toFVB/NJ mice (The Jackson Laboratory; strain #001800). Mice of eithersex were included in the analysis. All animal procedures were ap-proved and performed according to the guidelines of the InstitutionalAnimal Care and Use Committee of the University of Alabama atBirmingham and in accordance with the policies on the Use of Ani-mals and Humans in Neuroscience Research, revised and approved bythe Society of Neuroscience.

EEG acquisition. Intracranial recording electrodes (Plastics One) wereinserted through the cranium over each hemisphere, with a referenceelectrode placed over the left. Stainless steel screws (Small Parts) wereanchored into the skull to provide stability for the dental acrylic, appliedto cover the electrodes and the screws. The scalp incision was then closedwith skin glue. Animals were allowed to recover for 5 d before EEGrecording. For wired EEG recordings, animals were housed individuallyin eight specially constructed cages, each fitted with an amplifier(BIOPAC Systems) and a varifocal IR digital camera (Digimerge Tech-nology). EEG data were collected at a sampling rate of 500 Hz. Data wereacquired, digitized, and analyzed off-line using AcqKnowledge 4.1 Soft-ware (BIOPAC Systems). EEG data analysis, including duration, ampli-tude, and frequency of abnormal EEG events, was conducted by aninvestigator blinded to genotype and treatment groups. None of theanalyses was automated. Seizure data were collected from all of the datasampled. For animals with high interictal spiking frequencies, e.g., the�6 months of age group, several 1 h time frames were chosen randomly

from the entire recording. Time frames with noise distorting the EEGsignal frequently were excluded from the analysis of interictal spiking.Corresponding video recordings (L20WD800 Series; Lorex Technology),aligned temporally with EEG acquisition, were used to assess animalbehavior during abnormal EEG activity.

Whole-cell patch-clamp recordings. Animals were decapitated and thebrains removed and immersed in ice-cold cutting solution containingthe following (in mM): 135 N-methyl-D-gluconate, 1.5 KCl, 1.5 KH2PO4,23 choline bicarbonate, 25 D-glucose, 0.5 CaCl2, and 3.5 MgSO4. Coronalslices (300 �m) recovered 1 h in normal recording solution contained thefollowing (in mM): 125 NaCl, 3 KCl, 1.25 NaH2PO4, 25 NaHCO3, 2CaCl2, 1.3 or 2 MgSO4, and 25 D-glucose at 28°C and 1 h at room tem-perature before recording. Whole-cell patch-clamp recordings were ob-tained using an intracellular solution consisting of the following (in mM):134 K-gluconate, 1 KCl, 10 HEPES, 2 Mg-ATP, 0.2 Na-GTP, and 0.5EGTA, pH 7.4, 285–290 mOsm. Tight seals were established using patchelectrodes (KG-33 glass; Garner Glass) with a 3– 8 M� open-pipetteresistance at �70 mV for pyramidal neurons and �80 mV for astrocytesas previously described (Olsen et al., 2006). Layer II/III pyramidal cellsand astrocytes were visualized using a Zeiss Axio Scope microscope, a40 � water-immersion objective, and infrared or, alternatively in addi-tion, epifluorescent illumination (GFP-positive astrocytes in Aldh1l1-eGFP mice). Drugs were added to the recording solution (50 �M D-AP5,500 nM TTX, 0.1 mM CdCl2 (Sigma), 0.1 mM BaCl2 (Sigma), 20 �M

bicuculline methiodide, 0.3 mM DL-threo-�-benzyloxyaspartic acid(TBOA), and 0.4 mM dihydrokainic acid (DHK). Unless stated other-wise, all drugs were purchased from Tocris Bioscience. Data were ac-quired using Clampex 10 software and a 19 Digidata 1440A interface(Molecular Devices), filtered at 5 kHz, digitized at 10 –20 kHz, and ana-lyzed using Clampfit 10.0 software (Molecular Devices). Resting mem-brane potentials (RMPs; in I � 0 mode), whole-cell capacitance, andseries resistances were measured directly from the amplifier, �1 minafter whole-cell access was obtained. Series resistance compensation wasadjusted to 85% to reduce voltage errors. Slices were continuously per-fused with ACSF at 32°C.

Pressure application of potassium or glutamate was performed usingthe Pico-liter Injector PLI-10 (Warner Instruments) at an injection pres-sure of 2 psi for 50 ms (K � uptake) or 500 ms (glutamate uptake). Uptakewas recorded three times for each cell and condition. Traces were aver-aged before analysis.

Neocortical layer II/III pyramidal neurons from �1-integrin fl/fl,Nex::Cre �/� and �1-integrin fl/wt, Nex::Cre �/� mice were recorded us-ing a switched current- and voltage-clamp amplifier (SEC 10 L; npi).Recordings were performed using either the current-clamp mode or theswitched voltage-clamp mode at a switching frequency of �42 kHz (dutycycle 25%). Initially, the membrane potential and current–voltage rela-tionship of the neurons was assessed in the current-clamp mode. Furtherrecordings were made in voltage-clamp mode at a holding potential of�70 mV. Following a 10 min baseline recording, the normal recordingsolution was replaced by an Mg 2�-free solution. In some cases D-AP5 (10�M) or bicuculline methiodide (10 �M) was added to the Mg 2�-freebathing solution. Occasionally, neurons were recorded in current-clampmode following Mg 2� withdrawal. The recorded signals were filtered at10 kHz (current-clamp mode) or 5 kHz (voltage-clamp mode). For off-line analysis, the signals were sampled at a rate of 5–10 kHz using acomputer equipped with an analog/digital-converter (PCI-6024E; Na-tional Instruments). Sampling and data storage were controlled using theprogram CellWorks (npi). Data were analyzed using IGOR Pro software(WaveMetrics).

Histological procedures. Adult animals were deeply anesthetized withketamine/xylazine and transcardially perfused with PBS followed by 4%PFA (Santa Cruz Biotechnology; 50 ml/animal). Brains were postfixed inthe same fixative for 12 h at 4°C, washed in PBS, and embedded in 4%agarose for cutting 50 –100 �m vibratome sections. For immunofluores-cent labeling, sections were incubated overnight at 4°C in PBS containingthe first antibody, 0.5% Triton X-100 (TX), and 10% NGS, then washedin PBS and incubated for 1 h at room temperature (RT) in 0.5% TX and10% NGS containing the secondary antibody raised against mouse IgGor rabbit generated in goat and coupled to either Alexa 488, Alexa 550, or

Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy J. Neurosci., February 25, 2015 • 35(8):3330 –3345 • 3331

Alexa 647 (Jackson ImmunoResearch). Primary antibodies used in-cluded mouse anti-GFAP (Millipore; 1:1000), rabbit anti-GFAP (Dako,#Z0334; 1:1000), chicken anti-GFAP (Millipore; 1:1000), rat anti-Cd11b(AbD; Serotec, #MCA711; 1:500), rabbit anti-Kir4.1 (Alomone, #APC-035; 1:500), rabbit anti-glutamine synthetase (Abcam, #ab16802; 1:500),rabbit anti-Glt-1 (anti-B12, rat Glt-1 residues 12–26; batch number“ab#360,” 0.2 mg/ml; 1:1000; Holmseth et al., 2005). After washing inPBS, sections were mounted on glass slides, embedded in Aqua-Poly/Mount, and covered by a glass coverslip.

Quantification was performed using Fiji/ImageJ in confocal images ofsix different regions of interest within the cortical gray matter from threedifferent slices representing different planes along the anterior–posterioraxis of the forebrain and spanning all six cortical layers. After collapsingthe hyperstacks, binary images were made and the area occupied byGFAP immunohistochemistry was measured.

Western blotting. Anesthetized animals were decapitated and the cor-tical gray matter was dissected in ice-cold HBSS containing 10 mM

HEPES. Tissue was flash frozen on dry ice and stored at �80°C untiltissue lysis was performed. RIPA buffer (containing 50 mM TrisHCl, pH7.5, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, and 1% SDS)was supplemented with protease (Sigma) and phosphatase inhibitor(Sigma). This lysis buffer was added to the tissue samples and the tissuehomogenized using a Dounce tissue grinder (Wheaton). After sonication,homogenates were centrifuged for 5 min at 12,000 � g at 4°C. Protein quan-tification was performed on the supernatant using a Lowry protein assay kit(Bio-Rad) or a BCA protein assay kit (Thermo Scientific). Laemmli-SDSsample buffer, containing 600 mM �-mercaptoethanol, was added to theprotein samples before incubating them at 60°C for 15 min or at 95°C for10 min. Fifteen to twenty-five micrograms of protein were loaded intoeach lane of a 4 –15% gradient precast acrylamide SDS-PAGE gel (Bio-Rad). Samples were separated at 100 –200 V. Protein was transferred onto0.45 �m PVDF Transfer membranes (Thermo Scientific; #88518) inMini Trans-Blot Electrophoretic Transfer Cells (Bio-Rad) at 100 V for1h. Membranes were blocked in TBST (0.1% Tween 20 in TBS) containing10% nonfat dried milk for 30 min at room temperature or at 4°C over-night. Blots were incubated in primary antibody solution (TBST, 10%goat serum and antibody) for 2 h at room temperature or at 4°C over-night. Blots were washed in TBST 3� for 10 min and incubated insecondary anti-rabbit, anti-mouse, or anti-sheep horseradish perox-idase-conjugated antibody 1:1500 (Santa Cruz Biotechnology) for 1 h atroom temperature. Membranes were washed in TBST 3� for 10 min anddeveloped using Luminol (Santa Cruz Biotechnology) or Femto(Thermo Scientific) chemiluminescence kits, and chemiluminescencewas detected using a 4000 MM Kodak imaging station and Kodak mo-lecular imaging software v.4.0.4. Primary antibodies included mouseanti-GAPDH (Abcam, #ab9483; 1:5000), sheep anti-GLAST (anti A1, ratGLAST residues 1–25, batch #ab286; 1.2 mg/ml; 1:6000; Li et al., 2012),rabbit anti-Glt-1 (anti-B12, rat Glt-1 residues 12–26, batch #ab360; 0.2mg/ml; 1:3500; Holmseth et al., 2005), rabbit anti-Kir4.1 (Alomone Labs,#APC-035; 1:2500), and rabbit anti-GFAP (Dako, #Z00334; 1:10,000).

Statistical analysis. Statistics were computed, generated, and graphedwith GraphPad Prism 5 or 6 (GraphPad Software). The significance levelwas set to p 0.05. Data were tested for normal distribution using theKolmogorov–Smirnov normality test. Statistical tests were chosen ac-cordingly and are specified in the results. Unless stated otherwise, allvalues are reported as mean SEM, with n being the number of animalsor cells sampled, as detailed in Results. Box-and-whisker plots displayminimum to maximum values, and mean is shown as “�”. Biophysicalrecordings were plotted using Origin 8 (MicroCal) or IGOR Pro 6(WaveMetrics). Statistical significance with *p 0.05, **p 0.01, and***p 0.001.

ResultsConditional deletion of �1-integrin causes progressivereactive astrogliosisLoss of �1-integrins in radial glial progeny during the first post-natal week leads to widespread reactive astrogliosis in adult mice

without causing abnormal cortical layering, cell death, or open-ing of the blood– brain barrier (Robel et al., 2009).

We previously showed a relatively slow turnover of �1-integrins after Cre-mediated recombination �E13. Decreasedprotein levels were first observed at P0 and were further decreased1 week after birth (Robel et al., 2009). In the prior study weaddressed the consequences of loss of �1-integrins exclusively inadult mice. Here, we set out to determine the phenotype onset.We collected cortical gray matter tissue from �1�/� and controlmice at different ages for immunohistochemical and Westernblot analysis. While GFAP is expressed by radial glia late duringembryonic development in the healthy mouse brain, this inter-mediate filament is downregulated in most astrocytes of the cor-tical gray matter during the first 2 weeks after birth (Cahoy et al.,2008). Accordingly, control (�1-integrin w/w, hGFAP-Cre�; �1-integrin fl/fl, hGFAP-Cre-; �1-integrin fl/wt, hGFAP-Cre�) and�1-integrin fl/fl, hGFAP-Cre� (�1�/�) brains had similarly lownumbers of GFAP� astrocytes at 2 weeks of age (Fig. 1a). How-ever, at 4 weeks, GFAP was upregulated in �1�/� brains com-pared with littermate controls. GFAP� astrocytes were foundmainly in layer I/II and V/VI at 4 weeks, whereas at 6 weeks,almost all cortical gray matter astrocytes showed increased GFAPexpression (Fig. 1a; Robel et al., 2009). Western blot analysis ofcortical gray matter tissue confirmed significantly higher GFAPprotein levels between control and �1�/� cortical gray matter at4 and 6 weeks, and in tissue of animals older than 6 months (Fig.1b). Note that the quantification of the GFAP immunohisto-chemistry (Fig. 1a) represents the area occupied by GFAP� cellsrather than protein expression levels (Fig. 1b).

Astrocytic hypertrophy increased concomitantly with GFAPlevels (Fig. 1a, higher power images). We showed previously that,in addition, characteristic antibody markers for immatureastrocyte-like stem cells, such as vimentin and tenascin-C, areupregulated. However, in these animals, astrocytes do not un-dergo proliferation in vivo or in vitro and astrocyte densities areunchanged (Robel et al., 2009).

We previously observed mild microglia activation in adultmice that was unassociated with significantly increased microglianumbers or proliferation (Robel et al., 2009). Immunohisto-chemistry done in �1�/� mice at younger ages revealed mildmicroglial activation, including increased Cd11b expression anda deramified phenotype starting at �6 weeks of age (Fig. 1c).Interestingly, no morphological changes or difference in Cd11bimmunohistochemistry were observed at 4 weeks of age. ThusCd11b� reactive microglia appear later than GFAP� reactiveastrocytes, suggesting that microglia are activated as a conse-quence of astrocyte activation (see also (Robel et al., 2009).

Together, �1�/� mice display a progressive astrogliosis phe-notype with late postnatal onset and mild microgliosis secondaryto astrocyte activation.

Mice with chronic astrogliosis develop spontaneous seizuresEpileptic foci are known to contain reactive astrocytes (D’Ambrosio,2004; Wetherington et al., 2008; Carmignoto and Haydon, 2012;Coulter and Eid, 2012), yet whether they are a consequence ofneuronal dysfunction or an active contributor is currently un-known (Verkhratsky et al., 2012). Since hippocampal neuronalhyperexcitability had been observed following adenoviral-inducedastrogliosis (Ortinski et al., 2010), we sought to investigate whetherthe widespread astrogliosis in �1�/� mice leads to spontaneous sei-zure development.

To assess gross cortical activity, we recorded subcranial corti-cal EEGs from different age groups of �1�/� mice and controls.

3332 • J. Neurosci., February 25, 2015 • 35(8):3330 –3345 Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy

EEGs (22–258 h continuous recording) were recorded while si-multaneously videotaping each animal (see Table 1 for a list of all�1�/� animals and respective recording times).

We observed highly abnormal EEG activity in �1�/� micecompared with their control littermates (Fig. 2a). This activityconsisted of tonic– clonic seizures accompanied by a clear con-vulsive phenotype (Fig. 2b,d) and frequent interictal spiking (Fig.2c,e). Tonic– clonic seizures were exhibited by 58% of �1�/�

animals (29/50 mice), while seizures were never observed in con-trol littermates (0/48 mice). We recorded EEGs in four differentage groups (4 and 5– 6 weeks, 2–3 months, and older than 6

months), and found that the seizure incidence was highest in 5- to6-weeks-old animals (Fig. 2f). However, seizure frequency wasnot significantly different between age groups (ROUT outliertest, Kruskal—Wallis test, p � 0.297, Dunn’s multiple compari-son test; Fig. 2g). Given that seizures occurred on average onlyonce every 1–2 d, it is possible that the recording periods were notalways long enough to compute an accurate seizure frequency(Table 1). Thus, we might overestimate the number of animalswithout seizures.

In contrast, we observed age-dependent interictal spiking inalmost all �1�/� mice, which increased dramatically in older

Figure 1. Mice with deletion of �1-integrin (�1 �/�) develop widespread reactive astrogliosis and mild secondary microgliosis. a, Immunohistochemistry against the intermediate filamentGFAP in the cortical gray matter of �1-integrin-deficient and control mice at 2, 4, and 6 weeks and 3.5 months of age. GFAP levels are comparably low at 2 weeks of age, but progressively increasein �1 �/� mice of 4 and 6 weeks and 3.5 months of age. Cell bodies and main processes appear strongly hypertrophied in �1 �/� mice at 6 weeks of age and older (high magnification). In controls,GFAP-positive nonhypertrophied astrocytes can be found close to the meninges and surrounding large blood vessels (high magnification). Quantification of the area occupied by GFAP � cellsrevealed significant differences between control (n � 23) and �1 �/� mice older than 6 weeks (4 weeks of age, n � 3; 6 weeks of age, n � 9; 8 weeks of age, n � 10; �6 months of age, n �6; one-way ANOVA followed by Tukey’s multiple comparison test). b, Western blot analysis and quantification of GFAP at 4 and 6 weeks and �6 months of age. c, Immunohistochemistry againstthe microglial receptor Cd11b in �1 �/� and control mice at 2, 4, and 6 weeks of age. Staining intensity is increased in �1 �/� mice at 6 weeks of age indicating microglia activation. Error bars referto SEM. Statistical significance with *p 0.05, **p 0.01, and ***p 0.001.

Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy J. Neurosci., February 25, 2015 • 35(8):3330 –3345 • 3333

mice (ROUT outlier test, one-way ANOVA, p � 0.0001, Tukey’smultiple comparison test; Fig. 2h). Only 14% of �1�/� miceshowed none or little interictal spiking, whereas the remaining86% have considerably higher values (n � 43) when comparedwith littermate controls (n � 20). Half of the �1�/� mice (3/6)with little or no interictal spiking were 4 weeks of age, the agegroup presenting with less severe astrogliosis. This suggests a highpenetrance of the phenotype at least with regard to interictalspiking and GFAP expression (7% of the mice analyzed by im-munohistochemistry showed GFAP expression comparable to

controls while all other animals had elevated GFAP levels in thecortical gray matter, n � 28).

To assess a correlation between GFAP immunohistochemis-try and seizure frequency, the Spearman correlation coefficientbetween these two variables was computed. The coefficient of � �0.37 (p � 0.006) indicates that GFAP levels and seizure frequencytend to increase together.

We were unable to conduct continuous EEG recordings onmice younger than 4 weeks since animals are housed separatelyafter cranial electrode implantation, and therefore they had to beof weaning age for surgery. In addition, we waited �5 d afterimplantation before recording. We observed that �1�/� miceoften weakened and showed weight loss after several days of re-cording. These animals were removed from further recording.

It can be difficult to place EEG electrodes on the brain surfacewithout them penetrating into the brain or injuring the menin-ges, which can cause astrocyte and microglia activation. In a sub-set of control and �1�/� animals, we found that implantation ofthe electrodes had caused tissue damage (Fig. 3). To assess howthis acute injury affects seizures in control and �1�/� mice, weidentified the group of animals that had incurred additional dam-age based on GFAP and Cd11b immunohistochemistry. We di-vided the animals into three groups: animals without damage,animals with mild damage, or animals with severe damage. Inanimals without tissue damage we did not observe microgliaactivation as assessed by Cd11b levels (Fig. 3a). Note that themicroglia activation and the concomitant increase in Cd11b im-munoreactivity caused by electrode damage by far exceeds themild microglia activation that we normally observe in �1�/�

animals (Robel et al., 2009). This allowed us to identify electrode-injured brain areas despite high GFAP levels in �1�/� mice (Fig.3b). In mice with mild injury electrodes had not penetrated thebrain tissue but Cd11b (and GFAP in controls) was upregulatedwhen compared with neighboring brain areas. In animals withsevere damage, electrodes had penetrated the brain tissue.

We then compared seizure and interictal spiking frequenciesbetween animals with and without additional acute tissue dam-age. We never observed seizures in control animals. More specif-ically, 10 control animals with acute damage (Fig. 3c, mild andsevere) were recorded for a total of 769 h and six of these micewere recorded again 4 – 6 weeks later for a total of 331 h. The lackof seizures in control mice suggests that inflammation induced bythis acute and locally confined injury was not sufficient to triggerepileptogenesis at least during our observation period of up to 8weeks after electrode implantation.

To ensure that the acute injury and chronic astrogliosis in�1�/� mice do not have a synergistic effect on seizure and spikingfrequency, we compared �1�/� with and without electrode dam-age. There was no significant difference in interictal spiking orseizure frequencies between these groups (Fig. 3c).

These data show that chronic astrogliosis in �1�/� mice isassociated with abnormal EEG activity representative of interictalspiking and spontaneous seizures in a majority of mice whilefocal gliosis resulting from a localized acute injury does not causeepilepsy.

Neuronal hyperexcitability characterizes �1 �/� miceEpileptic seizures result from abnormal changes in neuronal ex-citability (Goldberg and Coulter, 2013). To examine whether thiswas the case in �1�/� animals, we performed whole-cell patch-clamp recordings on layer II/III cortical pyramidal neurons inacute brain slices of mice 4 – 6 and 6 – 8 weeks old. We found that

Table 1. Time span of EEG recordings, seizure number, and seizure frequency arelisted for EEG-recorded �1 �/� mice

Animal tag Recorded time (h) # of seizures Seizure frequency

4 weeks of age1689 53.58 0 0.0001696 53.58 0 0.0001698 22 0 0.0001700 53.58 1 0.0193125 163 0 0.000No ET 1 163 0 0.000No ET 2 163 2 0.0123165 257.93 24 0.0933166 257.93 11 0.043

6 weeks of age714 65.78 7 0.106704 65.78 9 0.137705 65.78 7 0.106707 65.78 7 0.1061851 69.55 0 0.0001852 69.55 0 0.0001854 69.55 0 0.0001857 39.95 0 0.0001864 69.55 2 0.0292152 78 4 0.0512159 78 5 0.0642175 78 2 0.0262176 78 1 0.0132171 78 1 0.013

2–3 months of age1614 52.2 13 0.2491615 52.2 5 0.0961623 52.2 0 0.0001630 52.2 0 0.0002230 67 2 0.0302208 67 0 0.0002210 67 0 0.000188 77.46 2 0.026201 88.36 2 0.023236 67.86 0 0.000239 69.6 0 0.000578 99.44 1 0.010579 99.44 1 0.010581 99.44 4 0.040582 99.44 3 0.030

�6 months of age256 52 12 0.231265 52 0 0.000268 52 0 0.000818 69.5 5 0.072866 69.5 1 0.014909 69.5 0 0.000911 69.5 3 0.043917 69.5 0 0.000923 69.5 5 0.072

EEGs were recorded from a corresponding number of control animals in each age group.

3334 • J. Neurosci., February 25, 2015 • 35(8):3330 –3345 Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy

the neuronal RMP was significantly more depolarized in�1 �/� neurons (4 – 6 weeks: RMP � �63.28 1.77 mV, n �32; 6 – 8 weeks: RMP � �68.87 1.38 mV, n � 30) than incontrols (4 – 6 weeks: RMP � �71.83 0.8 mV, n � 35,two-tailed unpaired t test, p 0.0001; 6 – 8 weeks: RMP ��73.03 0.77, n � 32, two-tailed unpaired t test, p � 0.0002;Fig. 4a), while the input resistances were higher in 4- to6-week-old �1 �/� mice (control Ri � 102.6 8.28 M�,�1 �/� Ri � 137.9.36 M�, two-tailed Mann–Whitney test, p �0.0014; Fig. 4b) but comparable between both groups at 6 – 8weeks of age (control Ri � 78.38 7.97 M�, �1 �/� Ri �84.86 5.55 M�, two-tailed Mann–Whitney test, p � 0.1016;Fig. 4b).

To examine whether �1 �/� pyramidal neurons are hyper-excitable, we chose four different approaches. First, we deter-mined the latency of the development of ictal (defined asevents longer than 2 s) and of interictal (defined as events with

a duration between 500 ms and 2 s) activity following removalof Mg 2� from the bath solution allowing for the removal ofthe Mg 2� block from the NMDA receptor as previously de-scribed (Buckingham et al., 2011). In control mice, latency tothe first interictal event was 23.69 3.25 min at 4 – 6 weeks(n � 12) and 15.49 2.29 min (n � 14) at 6 – 8 weeks of age(Fig. 4c– e). The onset of ictal discharges was 58.24 8.69 min(n � 9) and 37.93 2.53 min (n � 9) at 6 – 8 weeks of age (Fig.4c– e). Some neurons did not respond with ictal dischargeswithin the recording period of 60 –90 min. Nonresponsiveneurons were excluded from the analysis since we could notdetermine the onset time. There were more nonrespondingneurons in control brain slices compared with slices obtainedfrom �1 �/� mice, suggesting that the mean value for ictalonset in control animals is likely underestimated. Pyramidalneurons in �1 �/� brain slices showed a significantly shorterlatency to onset of interictal (4 – 6 weeks: 12.04 2.82 min,

Figure 2. Mice with chronic astrogliosis develop spontaneous seizures and interictal spiking. a, EEG recordings reveal abnormal EEG activity in �1 �/� mice when compared with control mice.b, Abnormal activity in �1 �/� mice manifests in spontaneous tonic– clonic seizures or in interictal spiking (c). d, High temporal resolution of the start of a seizure. At the end of the seizure, EEGsignal depression is typically observed. e, High temporal resolution of an interictal spike. f, The percentage of seizing animals is lowest at 4 weeks of age and increases in older animals. g, �1 �/�

mice experience an average of one to two seizures within 48 h. Seizure frequency is not significantly different between age groups. h, Interictal spiking occurs at a high frequency in all age groupsand is significantly increased in �1 �/� mice older than 6 months. Error bars refer to SEM. Statistical significance with *p 0.05, **p 0.01, and ***p 0.001.

Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy J. Neurosci., February 25, 2015 • 35(8):3330 –3345 • 3335

n � 12, two-tailed unpaired Mann–Whitney test, p � 0.0015; 6 – 8 weeks:8.64 0.91 min, n � 10, two-tailed un-paired t test, p � 0.0243) and ictal (4 – 6weeks: 21.43 4.94 min, n � 10, two-tailed unpaired Mann–Whitney test,p � 0.0095; 6 – 8 weeks: 15.27 1.78min, n � 10, two-tailed unpaired t test,p � 0.0001) activity in both age groups(Fig. 4c– e).

Similarly, we determined the latency ofthe development of ictal and of interictalactivity following application of bicucull-ine, which is a GABAA receptor antagonistfacilitating excitatory synchronization(Gutnick et al., 1982). Whereas the onsetof interictal events was unchanged be-tween controls and �1�/� at 4 – 6 weeks ofage (Fig. 4f, control; 8.33 1.54 min,n � 10; �1�/�; 6.1 0.59 min, n � 14,two-tailed unpaired Mann–Whitney test,p � 0.54), the onset of ictal events wassignificantly faster in �1�/� pyramidalcells (control: 17.21 3.25 min, n � 8;�1�/�: 10.27 1.58 min, n � 15, two-tailed unpaired t test, p � 0.041).

Moreover, we quantified the number ofaction potentials fired in response to currentinjections. We found that this number wassignificantly greater in �1�/� pyramidalcells than in control neurons in response to180, 200, 220, and 240 pA current injection(Fig. 4g,h), indicating a lowered excitabilitythreshold in �1�/� neurons of both agegroups.

Last, we recorded input/output curvesin control and �1�/� cortical slices. Astimulus electrode was used to deliver in-crementally increasing stimuli to deepcortical layer IV, and postsynaptic cur-rents were recorded in the superficial cor-tical layers (II/III). Input/output curveswere constructed of the response area andamplitude as a function of stimulus inten-sity (Fig. 4i). Recordings from �1�/�

pyramidal cells were significantly largerthan controls. Increasing stimulation re-sulted in greater excitatory response areasin �1�/� slices compared with controls(Fig. 4i; two-way ANOVA followed by Si-dak’s multiple comparison test, n � 5 foreach genotype).

Fast synaptic excitation relevant to ep-ilepsy is largely mediated by AMPA recep-tors, and AMPA receptor antagonists canreduce epileptiform activity in animalmodels and human epilepsy (Rogawski,2013). To address whether AMPA receptorexpression was changed in �1�/� mice, weperformed a Western blot analysis for theGluR2/3 subunit. At 4 and 6 weeks of age,protein levels were not significantly changed between control and�1�/� mice. In older animals, Glu2/3 levels were slightly reduced

Figure 3. Tissue damage caused by electrode implantation does not affect the seizure phenotype. Immunohistochem-istry for Cd11b as a marker of reactive microglia (a) and GFAP in �1 �/� and control mice that had undergone electrodeimplantation and EEG recordings (b). c, Quantification of animals without tissue damage or mild and severe tissue damage.�1 �/� mice were divided in two groups based on whether or not tissue damage was detected. Seizure and interictalspiking frequencies were computed and did not differ significantly (seizures: with damage, n � 8; without damage, n �22; Mann–Whitney, p � 0.86; interictal spikes: with damage, n � 5; without damage, n � 11; Mann–Whitney, p � 0.48;animals �6 months were excluded from the analysis since the frequency of interictal spikes is significantly higher at thisage). Box-and-whisker plots display minimum to maximum values, and mean is shown as �.

3336 • J. Neurosci., February 25, 2015 • 35(8):3330 –3345 Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy

(Fig. 4j). This excludes an increase in GluR2/3 protein expressionlevels as a contributor for seizure generation in �1�/� mice.

Neuronal deletion of �1-integrins does not induceneuronal hyperexcitabilityTo exclude the possibility that neurons rather than astrocyteswere selectively affected by the �1-integrin deletion, we per-formed the same experiments in mice where �1-integrin deletion

was neuron specific. To mediate recombination of the floxed�1-integrin gene, �1 fl/fl mice were bred to Nex::Cre�/� mice,which express Cre under the control of regulatory sequences ofthe transcription factor Nex (Goebbels et al., 2006). This leads togenetic deletion in intermediate progenitors and neurons of thedorsal telencephalon at �E11.5 (Goebbels et al., 2006). We pre-viously reported that astrogliosis was absent in these mice, despitethe early downregulation of �1-integrin proteins (Robel et al.,

Figure 4. Layer II/III pyramidal neurons are hyperexcitable in mice with chronic astrogliosis. a, The RMP of Layer II/III pyramidal neurons is significantly depolarized in acute slices of �1 �/� micecompared with controls. b, The input resistance differs at 4 weeks of age but is unchanged in �1 �/� neurons compared with controls. Whiskers in box-and-whisker plots show minimum tomaximum values. The middle line is plotted at the median. c, Whole-cell patch-clamp recordings from pyramidal neurons before (ACSF) and after removal of Mg 2� (Mg 2�-free) followed by AP-5shows a faster onset and longer duration of ictal (�2 s in duration) events in acute slices of �1 �/� mice compared with controls at 6 – 8 weeks of age. d, Quantification of the latency to onset ofictal events (defined as events of �2 s in length) after removal of Mg 2� in �1 �/� mice and littermate controls, as well as in �1 �/� Nex::Cre mice and their littermate controls. e, Quantificationof the latency to onset of interictal (defined as events of 500 ms to 2 s length) after removal of Mg 2�. f, Quantification of the latency to onset of interictal and ictal events after bicuculline application.g, Examples of voltage responses to increasing amplitude current injections, from �100 to �240 pA (in 20 pA steps) in whole-cell, current-clamped pyramidal peritumoral neurons in �1 �/� andcontrols. h, The average action potential number obtained in response to �20 to 240 pA current pulses was plotted as a function of applied current yielding the input– output curves shown. i,Graphical display of input/output curves of postsynaptic potentials recorded by whole-cell patch-clamp recording from pyramidal cells in superficial cortical layer II/III and the response area as afunction of stimulation intensity in b1 �/� and control cortical slices. j, Western blot of the GluR2/3 subunit of the AMPA receptor. Error bars refer to SEM. Box-and-whisker plots display minimumto maximum values, and mean is shown as �. Statistical significance with *p 0.05, **p 0.01, and ***p 0.001.

Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy J. Neurosci., February 25, 2015 • 35(8):3330 –3345 • 3337

2009). However, layer II/III pyramidal neuron whole-cell patch-clamp recordings conducted in acute brain slices from �1-integrinfl/fl, Nex::Cre�/� mice and their control littermates(�1-integrinfl/wt, Nex::Cre�/�; �1-integrinfl/fl, Nex::Cre�/�; �1-in-tegrinfl/wt, Nex::Cre�/�) showed that there is a similar latency toepileptiform event onset (longer than 2 s) in all genotypes followingthe removal of Mg2� (Fig. 4d; control, 34.21 10.51 min, n � 6;�1-integrinfl/fl, Nex::Cre�/� 40.0 4.52 min, n � 7, two-tailedunpaired t test, p � 0.6035).

Since neuron-specific �1-integrin deletion did not induce hy-perexcitability, we conclude that astrogliosis causes �1�/� pyra-midal cells to become depolarized and hyperexcitable.

Do reactive astrocytes in �1 �/� mice have impairedK � homeostasis?To elucidate possible mechanisms underlying neuronal hyperex-citability and seizures in �1�/� mice, we first focused on the wellestablished, astrocytic homeostatic functions. It has been sug-gested that impairments in astrocytic potassium and glutamatehomeostasis can contribute to neuronal hyperexcitability andseizure generation (Seifert et al., 2010). Thus, we used a combi-nation of Western blot, immunohistochemistry, and electro-physiological recordings to assess whether potassium andglutamate uptake is intact in �1�/� astrocytes.

Whole-cell astrocyte patch-clamp recordings were performedon acute brain slices obtained from �1�/� or control mice at 6 – 8weeks of age. To identify astrocytes, these animals expressedeGFP under the control of the astrocyte-specific Aldh1l1 pro-moter (Cahoy et al., 2008; Fig. 5a). The mean input resistance(control Ri � 18.45 1.44 M�, n � 57; �1�/� Ri � 22.01 2.84M�, n � 39; two-tailed Mann–Whitney test, p � 0.9985; Fig. 5b)and the mean RMP (control RMP � �78.08 0.47 mV, n � 32;�1�/� RMP � �78.19 0.50 mV, n � 30; two-tailed t test, p �0.8713; Fig. 5c), were comparable between �1�/� and controlastrocytes. However, cell capacitance, which correlates with cellsize, was significantly increased in �1�/� astrocytes (control C �10.54 0.91 pF, n � 45; �1�/� C � 12.98 0.87 pF, n � 42;two-tailed Mann–Whitney test, p � 0.0158; Fig. 5d), consistentwith their hypertrophied phenotype.

Astrocytic potassium (K�) conductance and uptake isthought to be mediated by the barium-sensitive potassium chan-nel Kir4.1 (Olsen and Sontheimer, 2008). To activate Kir4.1-mediated currents in these cells, we used a specific voltage-stepprotocol as previously described (Olsen et al., 2007), and isolatedthese currents using 100 �M Ba 2�, which specifically inhibits Kirchannels at this concentration (Ransom and Sontheimer, 1995).A point-by-point subtraction of the corresponding whole-cellcurrent was done to isolate the Ba2�-sensitive component (Kir cur-rent; Fig. 5e). We quantified the Kir current, IKir (expressed in pA/pF), as the Ba2�-sensitive conductance at �140 mV, normalized tothe whole-cell capacitance from these recordings (Fig. 5e). IKir wassignificantly reduced in �1�/� astrocytes (IKir � �195.4 26.09pA/pF, n � 26) compared with controls (control IKir � �257.4 20.7pA/pF,n�31;two-tailedMann–Whitneytest,p�0.0019;Fig.5e).

To assess whether this reduction in Kir currents impairs K�

uptake, we challenged astrocytes in �1�/� and control acutebrain slices with pressure-applied K� (125 mM KCl), which inducesan inward current under voltage-clamp conditions. The number ofions that moved across the membrane can be determined from thecharge transfer, i.e., the area of the Ba2�-sensitive current induced bythe puff. Surprisingly, there was no difference between �1�/� andcontrol astrocytes in the number of K� ions moving across the

membrane (control: 1.08 0.24 fmol K�, n � 13; �1�/�: 1.29 0.23, n � 11, two-tailed unpaired t test, p � 0.5438; Fig. 5f).

We complemented these studies with immunohistochemicaland Western blot analyses to address whether reduced Kir cur-rents were a result of reduced levels of Kir4.1 protein. However,there were no significant differences in Kir4.1 protein expression(relative expression levels normalized to GAPDH) or distribu-tion between �1�/� and control animals at 4 and 6 weeks (Fig.5g,h). There were, however, differences in levels of Kir4.1 expres-sion in control mouse parenchymal gray matter astrocytes. Mostastrocytes showed low Kir4.1 levels within the fine processes;however, some astrocytes expressed higher Kir4.1 levels in cellbodies and processes. A similar pattern could be seen in brainslices from �1�/� mice (Fig. 5g). Kir current reduction seems tooccur at a functional level and was not associated with reduced levelsof Kir4.1 protein. Interestingly, Kir4.1 protein levels were increasedin �1�/� animals �6 months old (Fig. 5h) possibly reflecting com-pensatory changes in �1�/� astrocytes at later stages.

In summary, reactive astrocytes in �1�/� mice show reducedKir currents despite normal Kir4.1 protein levels. However, thisdid not significantly alter the resting membrane potential or K�

uptake. The latter may be due to the large, poorly localized K�

challenge. A reduced density of Kir channels on fine processesmay, nevertheless, show impaired local K� homeostasis.

Do reactive astrocytes in �1 �/� mice show impairedglutamate homeostasis?To examine the possibility that glutamate uptake may be im-paired in �1�/� mice, we first analyzed expression of the twoastrocytic glutamate transporters, Glt-1 (slc1a2) and GLAST(slc1a3), by immunohistochemistry and Western blot. Levels ofboth proteins were similar at 4 and 6 weeks of age (Fig. 6a– c). In�1�/� animals older than 6 months, Glt-1 levels were slightlyreduced (Fig. 6b), whereas GLAST protein levels remained com-parable between controls and �1�/� mice at this age (Fig. 6b).There was no apparent difference in Glt-1 distribution betweenthe 6-week-old control and �1�/� mice (Fig. 6a; Lehre et al.,1995) on the differential expression of Glt-1 and GLAST).

To assess if glutamate transport is changed, whole-cell patch-clamp recordings from eGFP� astrocytes were performed inacute brain slices from �1�/� or control mice. High glutamateconcentrations (200 �M) were applied using a puff pipette, anduptake currents were recorded in �1�/� astrocytes and controlswhile blocking K� uptake with BaCl2, and neuronal activity withTTX, D- AP5, bicuculline, and CdCl2. Recordings were repeatedafter blocking glutamate transporters with the inhibitors TBOAand DHK (Fig. 6d, traces). The amount of glutamate taken upcomputed as the TBOA/DHK-sensitive charge transfer (area un-der the curve) and was significantly reduced in �1�/� astrocytes(Fig. 6d; control 0.0857 0.0123 fmol glutamate, n � 6; �1�/�

0.0204 0.0120, n � 5, two-tailed unpaired t test, p � 0.0045).Despite previous reports of a downregulation of the astrocytic

enzyme glutamine synthetase in reactive astrocytes (Ortinski etal., 2010), we found comparable levels of this protein in the cor-tical gray matter of �1�/� and control brains. We found glu-tamine synthetase (GS) protein levels to be significantly reducedin older mice of both genotypes (Fig. 6e).

These data suggest that impaired astrocytic uptake of gluta-mate may cause increased extracellular glutamate concentrationsthat could contribute to enhanced neuronal excitability.

3338 • J. Neurosci., February 25, 2015 • 35(8):3330 –3345 Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy

Chronic gliosis is associated with an immature expressionprofile of neuronal chloride transportersRecent studies have suggested that decreased inhibition, eitherfrom reduced GABA synthesis (Ortinski et al., 2010) or GABAacting as an excitatory instead of an inhibitory neurotransmitter,can contribute to neuronal hyperexcitability (Lee et al., 2011;Ben-Ari et al., 2012; Rangroo Thrane et al., 2013). The action ofGABA on neurons depends on neuronal intracellular chlorideconcentrations. Mature neurons normally have a low intracellu-lar chloride [Cl�]i concentration resulting in hyperpolarizingGABA responses (Kahle et al., 2008). Neuronal [Cl�]i concentra-

tions are maintained by the balanced expression of the cation-chloride cotransporters (CCC) NKCC1 and KCC2. Immatureneurons show high expression of NKCC1 relative to low KCC2expression, which maintains a high [Cl�]i. A developmental ex-pression level switch occurs during the second postnatal week inrodents. KCC2 levels increase and NKCC1 levels decrease, result-ing in low [Cl�]i, and hyperpolarized responses after GABAbinds to GABAA receptors (Kahle et al., 2008).

To determine whether CCC expression is changed in micewith chronic astrogliosis, we performed Western blots to detectKCC2 and NKCC1 (Fig. 7a) in �1�/� and control cortical gray

Figure 5. Potassium homeostasis is moderately impaired in mice with chronic astrogliosis. a, For whole-cell patch-clamp recordings, astrocytes were identified by their eGFP in acute slices of�1 �/� and control mice (6 – 8 weeks of age) carrying an allele expressing eGFP under the astrocyte-specific Aldh1l1 promoter. Placement of the recording pipette and puff pipette is indicated. b,The input resistance (Ri) and resting RMP (c) of parenchymal astrocytes in the cortical gray matter is unchanged between �1 �/� and control mice. d, The cell capacitance is increased in �1 �/�

astrocytes. e, To activate Kir4.1-mediated potassium currents in astrocytes a voltage-step protocol was used in ACSF and after bath application of the Kir4.1 inhibitor BaCl2. Traces represent apoint-by-point subtraction of the corresponding current traces isolating the Ba 2�-sensitive component of the whole-cell current. Quantification of Kir currents (IKir in pA/pF) as the Ba 2�-sensitiveconductance at �140 mV normalized to the whole-cell capacitance revealed significantly reduced Kir4.1 currents in �1 �/� astrocytes. f, Potassium uptake was similar in �1 �/� and controlastrocytes after pressure application of 125 mM KCl for 50 ms. Whiskers in box-and-whisker plots show minimum to maximum values. The line in the middle of the box is plotted at the median, �indicates the mean. g, The majority of astrocytes expresses Kir4.1 at lower levels in fine processes only. Few astrocytes express higher Kir4.1 levels in cell bodies and processes. h, Western blot analysisprobing for Kir4.1 protein showed similar levels of protein expression and distribution in the cortical gray matter of �1 �/� and control mice at 6 weeks of age, but increased levels in older mice.Error bars in bar graphs refer to SEM. Statistical significance with *p 0.05, **p 0.01, and ***p 0.001.

Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy J. Neurosci., February 25, 2015 • 35(8):3330 –3345 • 3339

matter (n � 4 for each genotype and age group). We found thatlevels of KCC2 and NKCC1 protein and the ratio between KCC2 andNKCC1 were unchanged at 4 weeks (Fig. 7a–d). At 6 weeks, KCC2levels remained unchanged, whereas NKCC1 expression was signif-icantly increased. In mice 6 months and older, KCC2 protein wassignificantly lower. In mice with chronic reactive astrogliosis, KCC2to NKCC1 ratios are decreased after they are 6 weeks old (Fig. 7a–d).

NKCC1 is also expressed in astrocytes and endothelial cells(Pedersen et al., 2006; Macaulay and Zeuthen, 2012). Hence, weused immunohistochemistry to detect NKCC1 in animals 3 monthsof age. In controls, NKCC1 was expressed evenly throughout thecortical gray matter sparing only neuronal nuclei. We attemptedto colocalize NKCC1 with fine astrocytic processes using Glt-1

(Fig. 6a) as a marker in confocal images. However, the resolutionachieved by light microscopy was insufficient to unequivocallydemonstrate an overlap between NKCC1 and Glt-1. Neverthe-less, in �1�/� mice, we found increased ring-like NKCC1 label-ing within the cell soma of NeuN-labeled neurons (Fig. 7e). Wedid not observe increased levels of NKCC1 in astrocyte cell bodiesor the main processes after labeling with GFAP (Fig. 7e).

Immunohistochemistry of the neuron-specific KCC2 re-vealed expression of this transporter at the periphery of the somaand within processes as previously reported (Campbell et al.,2015). In �1�/� animals the KCC2 expression pattern waschanged showing a more diffuse labeling not just at the peripherybut also within neuronal soma (Fig. 7f).

Figure 6. Glutamate uptake is impaired in mice with chronic astrogliosis. a, Immunohistochemistry against the glutamate transporter Glt-1 and Western blot probing for Glt-1(b) and GLAST inthe cortical gray matter of �1 �/� (c) and control mice at different ages (4 and 6 weeks and �6 months). d, Glutamate uptake was assessed after puff application of 200 �M glutamate for 500 msbefore and after the application of the glutamate transporter inhibitors TBOA and DHK. ACSF contained the inhibitors (inh.) TTX, AP-5, bicuculline, and CdCl2. TBOA/DHK-sensitive glutamate uptakewas reduced in �1 �/� astrocytes when compared with littermate controls. Whiskers in box-and-whisker plot show minimum to maximum values. The line in the middle of the box is plotted at themedian, � indicates the mean. e, Protein expression levels of the astrocytic enzyme glutamine synthetase was unchanged in the cortical gray matter of �1 �/� and control mice at 4 and 6 weeksof age, but reduced in both genotypes in mice older than 6 months. Error bars in bar graphs refer to SEM. Statistical significance with *p 0.05, **p 0.01, and ***p 0.001.

3340 • J. Neurosci., February 25, 2015 • 35(8):3330 –3345 Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy

Together, these data indicate that neuronal NKCC1 andKCC2 protein distribution and levels changed in �1�/� mice andthat their neurons have adopted an immature phenotype.

Can bumetanide treatment reverse the seizure phenotype?To study whether the increased expression of NKCC1 relative to KCC2in neurons is responsible for seizure activity (Kahle et al., 2008), weblocked NKCC1 in vivo with the diuretic drug bumetanide in �1�/�

and control mice (Fig. 8a,b). We recorded EEGs from 6- to 8-week-oldmice, startingwitha48–72hbaselinebefore treatment with a very lowdose of bumetanide (Rangroo Thrane et al., 2013; 5 mg/kg, i.p., twicedaily). We found that seizures were eliminated in a subset of �1�/�

animals (5/9), but the remaining animals did not respond (4/9; Fig.8a). GFAP levels determined by immunohistochemistry after com-

pletion of the EEG recordings did not differ between treated anduntreated animals (untreated �1�/�: 219,416 51,461 n � 9;bumetanide-treated: 152,292 66,867, n � 7, two-tailed Mann–Whitney, p � 0.46). We never observed seizures in control mice beforeor after bumetanide treatment.

DiscussionA new epilepsy animal model with spontaneousrecurrent seizuresThere is a strong association between epilepsy and astrogliosisin acute and chronic animal models, and in human epilepsy braintissue (D’Ambrosio, 2004; Wetherington et al., 2008; Carmignotoand Haydon, 2012; Coulter and Eid, 2012). However, it has beendifficult to study astrogliosis in epileptogenesis without con-

Figure 7. Bumetanide treatment rescues the seizure phenotype in a subset of mice with chronic astrogliosis. a, Western blot analysis of the potassium-chloride cotransporter (KCC2) and thesodium potassium-chloride cotransporter (NKCC1). b, Quantification of KCC2. c, NKCC1 expression levels after normalization to the loading control GAPDH. d, �1 �/� mice present with an upsetratio of the chloride transporters KCC2 and NKCC1 at 6 weeks and �6 months of age. Immunohistochemistry for NKCC (e) and KCC2 (f ) in �1 �/� and control cortical gray matter at 3 months of age.White arrow heads point at neurons with changed NKCC1 or KCC2 expression patterns. Blue arrow heads point at normal, ring-like expression of KCC2 along the periphery of neuronal cell bodies. Errorbars refer to SEM. Statistical significance with *p 0.05, **p 0.01, and ***p 0.001.

Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy J. Neurosci., February 25, 2015 • 35(8):3330 –3345 • 3341

founding factors such as damage to the blood– brain barrier, in-flammation, and pronounced cell death. Ortinski et al. (2010)reported hippocampal circuit hyperexcitability after induction ofastrogliosis, and also noted a reduction in inhibitory, but notexcitatory, synaptic currents. They found that the glutamate-glutamine cycle failed, resulting in a downregulation of the astro-cytic enzyme GS and insufficient GABA production. However, itis unknown whether astrogliosis and resulting hyperexcitabilityproduced recurrent seizures in vivo.

In our study, we report that genetically induced chronic astro-gliosis is sufficient to induce spontaneous, recurrent, generalizedtonic– clonic seizures in vivo. We observed astrocytic cellular andfunctional changes believed to contribute to seizure develop-ment, such as impaired Kir4.1 function and decreased glutamateuptake in acute slices from these mice. In addition, we foundindirect changes, such as mild microglial activation and achanged expression pattern of the cation-chloride cotransportersNKCC1 and KCC2 in cortical pyramidal neurons.

Epileptogenesis in the absence of blood– brain barrier defectsBlood– brain barrier breach has been reported in epilepsy animalmodels and in epilepsy patients (Heinemann et al., 2012; Janigro,2012). Serum factors are deposited in the brain parenchyma afterblood– brain barrier breakdown and this alone can generate in-flammation and astrogliosis and trigger seizures (Seiffert et al.,2004; Adams et al., 2007; Schachtrup et al., 2010; Heinemann etal., 2012; Janigro, 2012). One effect of albumin deposition is as-trocytic TGFRII activation, which can then lead to downregula-tion of Kir4.1 and Glt-1 through a TGF-�-mediated mechanism(Frigerio et al., 2012). It has been difficult to decipher whether

these serum factors impact neuronal excitability directly orindirectly.

In our model, astrogliosis occurs in the absence of blood–brain barrier damage (Robel et al., 2009), reflected by stable levelsof Kir4.1 and Glt-1 proteins and mild microglial activation dur-ing seizure onset. This suggests that astrogliosis by itself is suffi-cient to induce epileptogenesis.

Direct changes in reactive astrocytes contributeto epileptogenesisAstrocytes are essential for normal brain homeostasis. Theymaintain low concentrations of potassium and glutamate withinthe extracellular space and supply neurons with glutamine, a nec-essary precursor for glutamate and GABA synthesis. Yet in hu-man epileptic hippocampi, glutamate levels have been found tobe five times higher than in normal hippocampi (Coulter andEid, 2012).

In �1�/� mice with chronic astrogliosis, we found functionalimpairments in the Kir4.1 potassium channel and in glutamatetransporters, but their protein levels were unchanged during sei-zure onset. Glutamate transporter trafficking, required forproper glutamate uptake, is crucially dependent on the cytoskel-eton (Sullivan et al., 2007; Sheean et al., 2013). In reactive astro-cytes, the cytoskeleton is markedly changed (Robel et al., 2011).Thus, glutamate uptake could be disrupted due to improper lo-calization of the transporters despite normal total protein levels.

During development, immature, proliferating astrocytes donot yet express Kir4.1, and therefore have a depolarized RMP ofapproximately �50 mV (Ransom and Sontheimer, 1995; Bordeyand Sontheimer, 1997). As the brain matures and astrocytic

Figure 8. Bumetanide treatment reversed the seizure phenotype in a subset of mice with chronic astrogliosis. a, The NKCC1 inhibitor bumetanide reversed the seizure phenotype in a subset of�1 �/� mice, whereas another group of �1 �/� mice did not respond. b, EEG traces at different temporal resolution of a mouse (#1615) before treatment and within 2 h after bumetanidetreatment.

3342 • J. Neurosci., February 25, 2015 • 35(8):3330 –3345 Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy

Kir4.1 expression levels increase (Nwaobi et al., 2014), astrocytesacquire their characteristic, very negative, RMP of approximately�85 mV (Olsen and Sontheimer, 2008). At the same time, pro-toplasmic cortical gray matter astrocytes downregulate typicalproteins expressed in a dedifferentiated, immature state, such asGFAP (in mice), nestin, and vimentin or tenascin-C and 473HD,and exit the cell cycle. However, after adult brain injury, reactiveastrocytes re-adopt an immature phenotype and can re-enter thecell cycle, coinciding with the loss of Kir4.1 and cell membranedepolarization (MacFarlane and Sontheimer, 2000; Higashimoriand Sontheimer, 2007). Mice with chronic astrogliosis show up-regulation of some of these markers, but do not re-enter the cellcycle (Robel et al., 2009). Accordingly, we did not observe Kir4.1protein loss or membrane depolarization. Nonetheless, barium-sensitive Kir4.1-mediated potassium currents were reduced inmice with chronic astrogliosis. This alone may contribute to epi-leptogenesis since Kir4.1 conditional knock-out mice displaystress-induced seizures (Djukic et al., 2007).

Indirect effects of astrogliosis on the microenvironmentNeuronal chloride homeostasisIn the adult CNS, GABA is the main inhibitory neurotransmitterthat binds to the GABA A receptor. This chloride channel opensafter ligand binding, facilitating passive flow of chloride ions ei-ther into or out of the cell according to the equilibrium potential(Kahle et al., 2008).

During development, GABA acts as an excitatory stimulus forimmature neurons due to their high intracellular chloride con-centration. GABA binds to its receptor, the channel opens, andchloride flows out of the immature cell to stimulate membranedepolarization. During postnatal development, neuronal intra-cellular chloride concentrations decrease in response to an ex-pression switch of the CCC transporters, NKCC1 and KCC2.NKCC1 is a sodium potassium chloride cotransporter that im-ports chloride into the cell, whereas KCC2 is a potassium chloridetransporter that shunts chloride out of the cell. In the immatureneuron, the KCC2 to NKCC1 ratio is low, resulting in high intra-cellular chloride, while in the mature neuron, NKCC1 levels arelower and KCC2 expression is higher, resulting in low intracellu-lar chloride. Once GABA binds and the receptor opens, chlorideflows into the cell and hyperpolarizes the membrane (Kahle et al.,2008). Although a regulation of the KCC2 and NKCC1 transport-ers by protein phosphorylation has been demonstrated (Kaila etal., 2014), the signaling mechanisms that govern this develop-mental switch and hence the conversion of GABA from excitatoryto inhibitory transmitter remain largely elusive.

However, excitatory GABAergic signaling caused by per-turbed neuronal chloride transport has been previously observedin brain slices from temporal lobe epilepsy patients (Cohen et al.,2002). Since NKCC1 is also involved in cell volume regulation,this transporter is regularly induced after CNS injury (Cohen etal., 2002). For example, increased NKCC1 activity or expressionhas been reported following cerebral hypoxic–ischemic injury(Kahle et al., 2008), traumatic brain injury (Lu et al., 2008), andpilocarpine-induced status epilepticus (Brandt et al., 2010). During aseizure, glutamate receptor activation and high extracellular potas-sium activate NKCC1 via serine-threonine phosphorylation, possi-bly perpetuating abnormally high intracellular chloride levels inmature neurons (Kahle et al., 2008). Simultaneously, KCC2 can bedownregulated or functionally impaired after CNS injury, eitherby microglia-mediated BDNF release (Ferrini et al., 2013), orenhanced extracellular glutamate (Lee et al., 2011; Kahle et al.,2013), further shifting the KCC2 and NKCC1 ratio. This ratio

appears to be unaltered in 4-week-old �1�/� mice, but is shiftedin older animals, indicating that KCC2 protein downregulationand NKCC1 protein upregulation is secondary to astrogliosis.How gliosis induces changes in expression and localization ofthese transporters remains to be elucidated. Hence, the shift inNKCC1:KCC2 ratios are likely a consequence of seizures as pre-viously suggested (Brandt et al., 2010) further perpetuating thephenotype. In a subset of mice, treatment with the NKCC1 in-hibitor drug bumetanide alleviated the seizures.

NKCC1 is primarily a neuronal transporter although it is ex-pressed in astrocytes as well. Here it has been suggested to con-tribute to K� uptake and water uptake, which may affect the sizeof the extracellular space. In astrocytes, NKCC1 activation is as-sociated with elevated extracellular potassium concentrations,which can be associated with cellular swelling as a result of thecotransport of water (Macaulay and Zeuthen, 2012). A recentstudy clarified that NKCC1 did not contribute to potassiumclearance or extracellular space shrinkage after stimulation ofneuronal activity in hippocampal slices under physiological con-ditions (Larsen et al., 2014). Under pathological conditions,NKCC1 may contribute to astrocytic hypertrophy and edema(Rose and Karus, 2013). Thus, inhibiting NKCC1 in astrocytesmight have beneficial effects, but so far positive effects of NKCC1inhibition have largely been attributed to a reduction in edemaformation (Chen and Sun, 2005; Kahle et al., 2009; Rose andKarus, 2013).

In our studies, we do not observe edema formation in �1�/�

animals. While bumetanide might also affect astrocytic NKCC1,the here observed anti-seizure effect is most likely due to an in-hibition of neuronal Cl� accumulation.

Microglia activation— epilepsy and inflammationReactive astrocytes release cytokines that may induce changes intheir cell biology, gene, and protein expression (Sofroniew, 2009;Sofroniew and Vinters, 2010; Devinsky et al., 2013). Cytokinerelease also activates microglia, which stimulates more cytokinerelease from these cells, possibly inducing a feedback loop be-tween astrocytes and microglia (Liu et al., 2011). Indeed, weobserved a low degree of microglia activation secondary to astro-gliosis in 6-week-old �1�/� mice. This activation was very mild,with little Cd11b upregulation, morphological change, or signif-icant proliferation when compared with an acute injury (Robel etal., 2009). However, in �1�/� mice, activated microglia mightsignal back to astrocytes and reinforce astrogliosis.

This back and forth signaling may explain why astrogliosisand microgliosis become more severe as the animals age. We didnot observe an age-dependent increase in seizure frequency in�1�/� mice, but interictal spiking increased dramatically in miceolder than 6 months. Given the relatively low numbers of seizuresduring the recording period, computed seizure frequencies mightnot be accurate enough to detect an age-dependent difference inseizure frequency. Generally, astrocytic and neuronal factors in-volved in the maintenance of balanced excitation and inhibitionchanged in older mice. Repeated seizures might contribute to en-hanced inflammation at these later stages and this might perpetuatethe seizure phenotype as reported for mesial temporal lobe epilepsyand epileptic encephalopathy (Avanzini et al., 2013).

In summary, we present a mouse model of chronic astrogliosisin which animals develop spontaneous tonic– clonic seizures invivo generated by neuronal hyperexcitability. We identified twomechanisms that likely contribute to interictal spiking and sei-zures. First, impaired astrocytic glutamate uptake could drivehyperexcitability through the resulting increase in extracellular

Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy J. Neurosci., February 25, 2015 • 35(8):3330 –3345 • 3343

glutamate. Second, the reversed ratio of neuronal chloride trans-porters KCC2 and NKCC1 may disrupt the delicate balance ofexcitation and inhibition in these mice.

These data support the hypothesis that reactive astrogliosis issufficient to induce epileptic seizures.

ReferencesAdams RA, Schachtrup C, Davalos D, Tsigelny I, Akassoglou K (2007) Fibrin-

ogen signal transduction as a mediator and therapeutic target in inflamma-tion: lessons from multiple sclerosis. Curr Med Chem 14:2925–2936.CrossRef Medline

Avanzini G, Depaulis A, Tassinari A, de Curtis M (2013) Do seizures andepileptic activity worsen epilepsy and deteriorate cognitive function? Epi-lepsia 54 [Suppl 8]:14 –21. CrossRef Medline

Ben-Ari Y, Khalilov I, Kahle KT, Cherubini E (2012) The GABA excitatory/inhibitory shift in brain maturation and neurological disorders. Neuro-scientist 18:467– 486. CrossRef Medline

Bordey A, Sontheimer H (1997) Postnatal development of ionic currents inrat hippocampal astrocytes in situ. J Neurophysiol 78:461– 477. Medline

Bordey A, Sontheimer H (1998) Properties of human glial cells associatedwith epileptic seizure foci. Epilepsy Res 32:286 –303. CrossRef Medline

Bordey A, Lyons SA, Hablitz JJ, Sontheimer H (2001) Electrophysiologicalcharacteristics of reactive astrocytes in experimental cortical dysplasia.J Neurophysiol 85:1719 –1731. Medline

Brandt C, Nozadze M, Heuchert N, Rattka M, Loscher W (2010) Disease-modifying effects of phenobarbital and the NKCC1 inhibitor bumetanidein the pilocarpine model of temporal lobe epilepsy. J Neurosci 30:8602–8612. CrossRef Medline

Buckingham SC, Campbell SL, Haas BR, Montana V, Robel S, Ogunrinu T,Sontheimer H (2011) Glutamate release by primary brain tumors in-duces epileptic activity. Nat Med 17:1269 –1274. CrossRef Medline

Cahoy JD, Emery B, Kaushal A, Foo LC, Zamanian JL, Christopherson KS,Xing Y, Lubischer JL, Krieg PA, Krupenko SA, Thompson WJ, Barres BA(2008) A transcriptome database for astrocytes, neurons, and oligoden-drocytes: a new resource for understanding brain development and func-tion. J Neurosci 28:264 –278. CrossRef Medline

Campbell SL, Robel S, Cuddapah VA, Robert S, Buckingham SC, Kahle KT,Sontheimer H (2015) GABAergic disinhibition and impaired KCC2cotransporter activity underlie tumor-associated epilepsy. Glia 63:23–36.CrossRef Medline

Carmignoto G, Haydon PG (2012) Astrocyte calcium signaling and epi-lepsy. Glia 60:1227–1233. CrossRef Medline

Chen H, Sun D (2005) The role of Na-K-Cl co-transporter in cerebral isch-emia. Neurol Res 27:280 –286. CrossRef Medline

Cohen I, Navarro V, Clemenceau S, Baulac M, Miles R (2002) On the originof interictal activity in human temporal lobe epilepsy in vitro. Science298:1418 –1421. CrossRef Medline

Coulter DA, Eid T (2012) Astrocytic regulation of glutamate homeostasis inepilepsy. Glia 60:1215–1226. CrossRef Medline

D’Ambrosio R (2004) The role of glial membrane ion channels in seizuresand epileptogenesis. Pharmacol Ther 103:95–108. CrossRef Medline

Devinsky O, Vezzani A, Najjar S, De Lanerolle NC, Rogawski MA (2013)Glia and epilepsy: excitability and inflammation. Trends Neurosci 36:174 –184. CrossRef Medline

Djukic B, Casper KB, Philpot BD, Chin LS, McCarthy KD (2007) Condi-tional knock-out of Kir4.1 leads to glial membrane depolarization, inhi-bition of potassium and glutamate uptake, and enhanced short-termsynaptic potentiation. J Neurosci 27:11354 –11365. CrossRef Medline

Ferrini F, Trang T, Mattioli TA, Laffray S, Del’Guidice T, Lorenzo LE, Cas-tonguay A, Doyon N, Zhang W, Godin AG, Mohr D, Beggs S, Vandal K,Beaulieu JM, Cahill CM, Salter MW, De Koninck Y (2013) Morphinehyperalgesia gated through microglia-mediated disruption of neuronalCl(-) homeostasis. Nat Neurosci 16:183–192. CrossRef Medline

Frigerio F, Frasca A, Weissberg I, Parrella S, Friedman A, Vezzani A, Noe FM(2012) Long-lasting pro-ictogenic effects induced in vivo by rat brainexposure to serum albumin in the absence of concomitant pathology.Epilepsia 53:1887–1897. CrossRef Medline

Goebbels S, Bormuth I, Bode U, Hermanson O, Schwab MH, Nave KA(2006) Genetic targeting of principal neurons in neocortex and hip-pocampus of NEX-Cre mice. Genesis 44:611– 621. CrossRef Medline

Goldberg EM, Coulter DA (2013) Mechanisms of epileptogenesis: a conver-

gence on neural circuit dysfunction. Nat Rev Neurosci 14:337–349.CrossRef Medline

Gutnick MJ, Connors BW, Prince DA (1982) Mechanisms of neocorticalepileptogenesis in vitro. J Neurophysiol 48:1321–1335. Medline

Heinemann U, Kaufer D, Friedman A (2012) Blood-brain barrier dysfunc-tion, TGFbeta signaling, and astrocyte dysfunction in epilepsy. Glia 60:1251–1257. CrossRef Medline

Higashimori H, Sontheimer H (2007) Role of Kir4.1 channels in growthcontrol of glia. Glia 55:1668 –1679. CrossRef Medline

Holmseth S, Dehnes Y, Bjørnsen LP, Boulland JL, Furness DN, Bergles D,Danbolt NC (2005) Specificity of antibodies: unexpected cross-reactivity of antibodies directed against the excitatory amino acid trans-porter 3 (EAAT3). Neuroscience 136:649 – 660. CrossRef Medline

Janigro D (2012) Are you in or out? Leukocyte, ion, and neurotransmitter permea-bility across the epileptic blood-brain barrier. Epilepsia 53 [Suppl 1]:26–34.CrossRef Medline

Jefferys JG (1990) Basic mechanisms of focal epilepsies. Exp Physiol 75:127–162. CrossRef Medline

Kahle KT, Staley KJ, Nahed BV, Gamba G, Hebert SC, Lifton RP, Mount DB(2008) Roles of the cation-chloride cotransporters in neurological dis-ease. Nat Clin Pract Neurol 4:490 –503. CrossRef Medline

Kahle KT, Simard JM, Staley KJ, Nahed BV, Jones PS, Sun D (2009) Molec-ular mechanisms of ischemic cerebral edema: role of electroneutral iontransport. Physiology 24:257–265. CrossRef Medline

Kahle KT, Deeb TZ, Puskarjov M, Silayeva L, Liang B, Kaila K, Moss SJ(2013) Modulation of neuronal activity by phosphorylation of the K-Clcotransporter KCC2. Trends Neurosci 36:726 –737. CrossRef Medline

Kaila K, Price TJ, Payne JA, Puskarjov M, Voipio J (2014) Cation-chloridecotransporters in neuronal development, plasticity and disease. Nat RevNeurosci 15:637– 654. CrossRef Medline

Larsen BR, Assentoft M, Cotrina ML, Hua SZ, Nedergaard M, Kaila K, VoipioJ, MacAulay N (2014) Contributions of the Na(�)/K(�)-ATPase,NKCC1, and Kir4.1 to hippocampal K(�) clearance and volume re-sponses. Glia 62:608 – 622. CrossRef Medline

Lee HH, Deeb TZ, Walker JA, Davies PA, Moss SJ (2011) NMDA receptoractivity downregulates KCC2 resulting in depolarizing GABAA receptor-mediated currents. Nat Neurosci 14:736 –743. CrossRef Medline

Lehre KP, Levy LM, Ottersen OP, Storm-Mathisen J, Danbolt NC (1995)Differential expression of two glial glutamate transporters in the rat brain:quantitative and immunocytochemical observations. J Neurosci 15:1835–1853. Medline

Li Y, Zhou Y, Danbolt NC (2012) The rates of postmortem proteolysis ofglutamate transporters differ dramatically between cells and betweentransporter subtypes. J Histochem Cytochem 60:811– 821. CrossRefMedline

Liu W, Tang Y, Feng J (2011) Cross talk between activation of microglia andastrocytes in pathological conditions in the central nervous system. LifeSci 89:141–146. CrossRef Medline

Lu KT, Cheng NC, Wu CY, Yang YL (2008) NKCC1-mediated traumaticbrain injury-induced brain edema and neuron death via Raf/MEK/MAPKcascade. Crit Care Med 36:917–922. CrossRef Medline

Macaulay N, Zeuthen T (2012) Glial K(�) clearance and cell swelling: keyroles for cotransporters and pumps. Neurochem Res 37:2299 –2309.CrossRef Medline

MacFarlane SN, Sontheimer H (2000) Changes in ion channel expressionaccompany cell cycle progression of spinal cord astrocytes. Glia 30:39 – 48.CrossRef Medline

Nwaobi SE, Lin E, Peramsetty SR, Olsen ML (2014) DNA methylation func-tions as a critical regulator of Kir4.1 expression during CNS development.Glia 62:411– 427. CrossRef Medline

Olsen ML, Sontheimer H (2008) Functional implications for Kir4.1 chan-nels in glial biology: from K� buffering to cell differentiation. J Neuro-chem 107:589 – 601. CrossRef Medline

Olsen ML, Higashimori H, Campbell SL, Hablitz JJ, Sontheimer H (2006)Functional expression of Kir4.1 channels in spinal cord astrocytes. Glia53:516 –528. CrossRef Medline

Olsen ML, Campbell SL, Sontheimer H (2007) Differential distribution ofKir4.1 in spinal cord astrocytes suggests regional differences in K� ho-meostasis. J Neurophysiol 98:786 –793. CrossRef Medline

Ortinski PI, Dong J, Mungenast A, Yue C, Takano H, Watson DJ, Haydon PG,Coulter DA (2010) Selective induction of astrocytic gliosis generates

3344 • J. Neurosci., February 25, 2015 • 35(8):3330 –3345 Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy

deficits in neuronal inhibition. Nat Neurosci 13:584 –591. CrossRefMedline

Pedersen SF, O’Donnell ME, Anderson SE, Cala PM (2006) Physiology andpathophysiology of Na�/H� exchange and Na�-K�-2Cl � cotransportin the heart, brain, and blood. Am J Physiol Regul Integr Comp Physiol291:R1–R25. CrossRef Medline

Pinto DJ, Patrick SL, Huang WC, Connors BW (2005) Initiation, propaga-tion, and termination of epileptiform activity in rodent neocortex in vitroinvolve distinct mechanisms. J Neurosci 25:8131– 8140. CrossRefMedline

Potocnik AJ, Brakebusch C, Fassler R (2000) Fetal and adult hematopoieticstem cells require beta1 integrin function for colonizing fetal liver, spleen,and bone marrow. Immunity 12:653– 663. CrossRef Medline

Rangroo Thrane V, Thrane AS, Wang F, Cotrina ML, Smith NA, Chen M, XuQ, Kang N, Fujita T, Nagelhus EA, Nedergaard M (2013) Ammoniatriggers neuronal disinhibition and seizures by impairing astrocyte potas-sium buffering. Nat Med 19:1643–1648. CrossRef Medline

Ransom CB, Sontheimer H (1995) Biophysical and pharmacological char-acterization of inwardly rectifying K� currents in rat spinal cord astro-cytes. J Neurophysiol 73:333–346. Medline

Robel S, Mori T, Zoubaa S, Schlegel J, Sirko S, Faissner A, Goebbels S, DimouL, Gotz M (2009) Conditional deletion of beta1-integrin in astrogliacauses partial reactive gliosis. Glia 57:1630 –1647. CrossRef Medline

Robel S, Bardehle S, Lepier A, Brakebusch C, Gotz M (2011) Genetic dele-tion of cdc42 reveals a crucial role for astrocyte recruitment to the injurysite in vitro and in vivo. J Neurosci 31:12471–12482. CrossRef Medline

Rogawski MA (2013) AMPA receptors as a molecular target in epilepsytherapy. Acta Neurol Scand Suppl 9 –18. CrossRef Medline

Rose CR, Karus C (2013) Two sides of the same coin: sodium homeostasisand signaling in astrocytes under physiological and pathophysiologicalconditions. Glia 61:1191–1205. CrossRef Medline

Schachtrup C, Ryu JK, Helmrick MJ, Vagena E, Galanakis DK, Degen JL,Margolis RU, Akassoglou K (2010) Fibrinogen triggers astrocyte scarformation by promoting the availability of active TGF-beta after vasculardamage. J Neurosci 30:5843–5854. CrossRef Medline

Seifert G, Steinhauser C (2011) Neuron-astrocyte signaling and epilepsy.Exp Neurol 244:4 –10. CrossRef Medline

Seifert G, Carmignoto G, Steinhauser C (2010) Astrocyte dysfunction inepilepsy. Brain Res Rev 63:212–221. CrossRef Medline

Seiffert E, Dreier JP, Ivens S, Bechmann I, Tomkins O, Heinemann U, Fried-man A (2004) Lasting blood-brain barrier disruption induces epilepticfocus in the rat somatosensory cortex. J Neurosci 24:7829 –7836. CrossRefMedline

Sheean RK, Lau CL, Shin YS, O’Shea RD, Beart PM (2013) Links betweenl-glutamate transporters, Na(�)/K(�)-ATPase and cytoskeleton in as-trocytes: evidence following inhibition with rottlerin. Neuroscience 254:335–346. CrossRef Medline

Sofroniew MV (2009) Molecular dissection of reactive astrogliosis and glialscar formation. Trends Neurosci 32:638 – 647. CrossRef Medline

Sofroniew MV, Vinters HV (2010) Astrocytes: biology and pathology. ActaNeuropathol 119:7–35. CrossRef Medline

Sullivan SM, Lee A, Bjorkman ST, Miller SM, Sullivan RK, Poronnik P, Cold-itz PB, Pow DV (2007) Cytoskeletal anchoring of GLAST determinessusceptibility to brain damage: an identified role for GFAP. J Biol Chem282:29414 –29423. CrossRef Medline

Thurman DJ, Beghi E, Begley CE, Berg AT, Buchhalter JR, Ding D, HesdorfferDC, Hauser WA, Kazis L, Kobau R, Kroner B, Labiner D, Liow K, Logros-cino G, Medina MT, Newton CR, Parko K, Paschal A, Preux PM, SanderJW, et al. (2011) Standards for epidemiologic studies and surveillance ofepilepsy. Epilepsia 52 [Suppl 7]:2–26. CrossRef Medline

Trevelyan AJ, Sussillo D, Watson BO, Yuste R (2006) Modular propagationof epileptiform activity: evidence for an inhibitory veto in neocortex.J Neurosci 26:12447–12455. CrossRef Medline

Verkhratsky A, Sofroniew MV, Messing A, deLanerolle NC, Rempe D, Rodrí-guez JJ, Nedergaard M (2012) Neurological diseases as primary gliopa-thies: a reassessment of neurocentrism. ASN Neuro 4:e00082. CrossRefMedline

Volterra A, Meldolesi J (2005) Astrocytes, from brain glue to communica-tion elements: the revolution continues. Nat Rev Neurosci 6:626 – 640.CrossRef Medline

Wahab A, Albus K, Gabriel S, Heinemann U (2010) In search of models ofpharmacoresistant epilepsy. Epilepsia 51 [Suppl 3]:154 –159. CrossRefMedline

Wetherington J, Serrano G, Dingledine R (2008) Astrocytes in the epilepticbrain. Neuron 58:168 –178. CrossRef Medline

Zhuo L, Theis M, Alvarez-Maya I, Brenner M, Willecke K, Messing A (2001)hGFAP-cre transgenic mice for manipulation of glial and neuronal func-tion in vivo. Genesis 31:85–94. CrossRef Medline

Robel et al. • Astrogliosis Is Sufficient to Cause Epilepsy J. Neurosci., February 25, 2015 • 35(8):3330 –3345 • 3345